Repository navigation
What_is_Orrin
Orrin is an experimental symbolic-first cognitive runtime: a long-running Python agent whose core — goals, memory, attention, control signals, and action selection — is symbolic and inspectable, with the language model as one optional tool among many. It runs continuously on its own loop rather than waiting for prompts, and it exposes its internals as named rooms in a live UI instead of hiding behavior in a chat transcript.
Orrin is a research prototype: not production software and not a claim of sentience. Cognitive terms name engineering mechanisms — memory, goals, attention, control signals, runtime state, and action selection.
Make the language model the smallest part of the agent, not the whole of it. Orrin runs fully with no API key: its memory, goals, priority weights, control signals, and reasoning continue in symbolic-only mode. When an LLM is configured, it is treated as a gated tool — see Symbolic-First Design and LLM Integration.
- Continuous runtime — a heartbeat loop, not a request/response bot (The Cognitive Loop).
- A consciousness bottleneck — a global workspace + ignition gate decide what reaches "mind" and when to think hard (Workspace and Ignition).
- An affective layer — regulated control signals bias every decision (Control Signals).
- Grounded reward — it is paid for durable outward effects, not internal churn (Production and the Effect Ledger).
- A finite life — a persistent lifetime clock with phases and a terminal cycle (Existence and Lifecycle).
- Host coupling — the machine it runs on is part of its context (Host Coupling).
- Researchers exploring hybrid symbolic/neural architectures and long-running agent behavior.
- Developers building durable-state agents who want inspectable internals.
- Anyone curious what an agent looks like when the LLM is a tool, not the controller.
- New here → Getting Started
- Want the mechanisms → The Cognitive Loop, then the subsystem pages
- Just browsing → Quick Navigation Guide and the Glossary
- Cognitive Loop
- Workspace & Ignition
- Binding & Writeback
- Thinking & Action Selection
- Learning & Adaptation
- Effect Ledger
- Quality Standard
- Expression Membrane
- Self-Code
- Native Language Model
- Existence & Lifecycle